Build AI Like Software, Not Spaghetti
Remember when jQuery spaghetti became React components? When monoliths became microservices? It’s time for AI development to grow up. Package expertise, compose intelligence, ship deterministically.
Perstack brings modern software engineering to AI: version your prompts like npm packages, compose them like React components, and deploy with the confidence of Docker containers.
The Expert Stack
Perstack is the open source ecosystem that transforms autonomous agents into composable, versionable packages that developers can build, share, and scale like modern software components.
Powered by the Expert Stack, Perstack brings the same composability and modularity that transformed web development to AI agents. Just as React components changed UI development, Experts let you build intelligent apps the way you build software.
Learn more in the Expert Stack documentation.
Fun fact: Perstack comes from the Latin “perītus” meaning “expert”.
The Third Way: Modern Software Engineering for AI
The AI development world is split between two extremes:
Big Workflow
Building massive orchestration graphs that become the legacy enterprise service bus of AI. Complex, brittle, and impossible to maintain.
Big Model
Hoping the next generation model will magically solve everything. No structure, no reproducibility—just prompts and prayers.
Expert Stack
We brought the lessons of modern software engineering to AI:
- Composable experts instead of monolithic chains
- Isolated contexts instead of context window explosions
- Git-like checkpoints instead of unreproducible runs
- npm-style packages instead of copy-paste prompts
Key Features
- 📦 Packaged Experts: Define in
perstack.toml, publish to registry, import as dependencies - 🧩 Expert Composition: Experts delegate tasks to others for complex multi-agent autonomy
- 🔄 Deterministic Runtime: Traceable, resumable, and forkable execution via checkpoints
- 🚀 Ready for Production: Same config from CLI to production apps
- 🛠️ MCP-Native: Integrate any MCP server with versioned, declarative config
- 📝 Workspace: Isolated environments where experts share data, not conversation history
- 🌐 Community Experts: Discover, share, and version AI experts like npm packages